GNU Linux-libre 4.19.286-gnu1
[releases.git] / arch / s390 / kernel / smp.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  SMP related functions
4  *
5  *    Copyright IBM Corp. 1999, 2012
6  *    Author(s): Denis Joseph Barrow,
7  *               Martin Schwidefsky <schwidefsky@de.ibm.com>,
8  *               Heiko Carstens <heiko.carstens@de.ibm.com>,
9  *
10  *  based on other smp stuff by
11  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
12  *    (c) 1998 Ingo Molnar
13  *
14  * The code outside of smp.c uses logical cpu numbers, only smp.c does
15  * the translation of logical to physical cpu ids. All new code that
16  * operates on physical cpu numbers needs to go into smp.c.
17  */
18
19 #define KMSG_COMPONENT "cpu"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21
22 #include <linux/workqueue.h>
23 #include <linux/bootmem.h>
24 #include <linux/export.h>
25 #include <linux/init.h>
26 #include <linux/mm.h>
27 #include <linux/err.h>
28 #include <linux/spinlock.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/delay.h>
31 #include <linux/interrupt.h>
32 #include <linux/irqflags.h>
33 #include <linux/cpu.h>
34 #include <linux/slab.h>
35 #include <linux/sched/hotplug.h>
36 #include <linux/sched/task_stack.h>
37 #include <linux/crash_dump.h>
38 #include <linux/memblock.h>
39 #include <linux/kprobes.h>
40 #include <asm/asm-offsets.h>
41 #include <asm/diag.h>
42 #include <asm/switch_to.h>
43 #include <asm/facility.h>
44 #include <asm/ipl.h>
45 #include <asm/setup.h>
46 #include <asm/irq.h>
47 #include <asm/tlbflush.h>
48 #include <asm/vtimer.h>
49 #include <asm/lowcore.h>
50 #include <asm/sclp.h>
51 #include <asm/vdso.h>
52 #include <asm/debug.h>
53 #include <asm/os_info.h>
54 #include <asm/sigp.h>
55 #include <asm/idle.h>
56 #include <asm/nmi.h>
57 #include <asm/topology.h>
58 #include "entry.h"
59
60 enum {
61         ec_schedule = 0,
62         ec_call_function_single,
63         ec_stop_cpu,
64 };
65
66 enum {
67         CPU_STATE_STANDBY,
68         CPU_STATE_CONFIGURED,
69 };
70
71 static DEFINE_PER_CPU(struct cpu *, cpu_device);
72
73 struct pcpu {
74         struct lowcore *lowcore;        /* lowcore page(s) for the cpu */
75         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
76         unsigned long ec_clk;           /* sigp timestamp for ec_xxx */
77         signed char state;              /* physical cpu state */
78         signed char polarization;       /* physical polarization */
79         u16 address;                    /* physical cpu address */
80 };
81
82 static u8 boot_core_type;
83 static struct pcpu pcpu_devices[NR_CPUS];
84
85 unsigned int smp_cpu_mt_shift;
86 EXPORT_SYMBOL(smp_cpu_mt_shift);
87
88 unsigned int smp_cpu_mtid;
89 EXPORT_SYMBOL(smp_cpu_mtid);
90
91 #ifdef CONFIG_CRASH_DUMP
92 __vector128 __initdata boot_cpu_vector_save_area[__NUM_VXRS];
93 #endif
94
95 static unsigned int smp_max_threads __initdata = -1U;
96
97 static int __init early_nosmt(char *s)
98 {
99         smp_max_threads = 1;
100         return 0;
101 }
102 early_param("nosmt", early_nosmt);
103
104 static int __init early_smt(char *s)
105 {
106         get_option(&s, &smp_max_threads);
107         return 0;
108 }
109 early_param("smt", early_smt);
110
111 /*
112  * The smp_cpu_state_mutex must be held when changing the state or polarization
113  * member of a pcpu data structure within the pcpu_devices arreay.
114  */
115 DEFINE_MUTEX(smp_cpu_state_mutex);
116
117 /*
118  * Signal processor helper functions.
119  */
120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, unsigned long parm)
121 {
122         int cc;
123
124         while (1) {
125                 cc = __pcpu_sigp(addr, order, parm, NULL);
126                 if (cc != SIGP_CC_BUSY)
127                         return cc;
128                 cpu_relax();
129         }
130 }
131
132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
133 {
134         int cc, retry;
135
136         for (retry = 0; ; retry++) {
137                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
138                 if (cc != SIGP_CC_BUSY)
139                         break;
140                 if (retry >= 3)
141                         udelay(10);
142         }
143         return cc;
144 }
145
146 static inline int pcpu_stopped(struct pcpu *pcpu)
147 {
148         u32 uninitialized_var(status);
149
150         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
151                         0, &status) != SIGP_CC_STATUS_STORED)
152                 return 0;
153         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
154 }
155
156 static inline int pcpu_running(struct pcpu *pcpu)
157 {
158         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
159                         0, NULL) != SIGP_CC_STATUS_STORED)
160                 return 1;
161         /* Status stored condition code is equivalent to cpu not running. */
162         return 0;
163 }
164
165 /*
166  * Find struct pcpu by cpu address.
167  */
168 static struct pcpu *pcpu_find_address(const struct cpumask *mask, u16 address)
169 {
170         int cpu;
171
172         for_each_cpu(cpu, mask)
173                 if (pcpu_devices[cpu].address == address)
174                         return pcpu_devices + cpu;
175         return NULL;
176 }
177
178 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
179 {
180         int order;
181
182         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
183                 return;
184         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
185         pcpu->ec_clk = get_tod_clock_fast();
186         pcpu_sigp_retry(pcpu, order, 0);
187 }
188
189 #define ASYNC_FRAME_OFFSET (ASYNC_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
190 #define PANIC_FRAME_OFFSET (PAGE_SIZE - STACK_FRAME_OVERHEAD - __PT_SIZE)
191
192 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
193 {
194         unsigned long async_stack, panic_stack;
195         struct lowcore *lc;
196
197         if (pcpu != &pcpu_devices[0]) {
198                 pcpu->lowcore = (struct lowcore *)
199                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
200                 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
201                 panic_stack = __get_free_page(GFP_KERNEL);
202                 if (!pcpu->lowcore || !panic_stack || !async_stack)
203                         goto out;
204         } else {
205                 async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
206                 panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
207         }
208         lc = pcpu->lowcore;
209         memcpy(lc, &S390_lowcore, 512);
210         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
211         lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
212         lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
213         lc->cpu_nr = cpu;
214         lc->spinlock_lockval = arch_spin_lockval(cpu);
215         lc->spinlock_index = 0;
216         lc->br_r1_trampoline = 0x07f1;  /* br %r1 */
217         if (nmi_alloc_per_cpu(lc))
218                 goto out;
219         if (vdso_alloc_per_cpu(lc))
220                 goto out_mcesa;
221         lowcore_ptr[cpu] = lc;
222         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
223         return 0;
224
225 out_mcesa:
226         nmi_free_per_cpu(lc);
227 out:
228         if (pcpu != &pcpu_devices[0]) {
229                 free_page(panic_stack);
230                 free_pages(async_stack, ASYNC_ORDER);
231                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
232         }
233         return -ENOMEM;
234 }
235
236 #ifdef CONFIG_HOTPLUG_CPU
237
238 static void pcpu_free_lowcore(struct pcpu *pcpu)
239 {
240         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
241         lowcore_ptr[pcpu - pcpu_devices] = NULL;
242         vdso_free_per_cpu(pcpu->lowcore);
243         nmi_free_per_cpu(pcpu->lowcore);
244         if (pcpu == &pcpu_devices[0])
245                 return;
246         free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
247         free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
248         free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
249 }
250
251 #endif /* CONFIG_HOTPLUG_CPU */
252
253 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
254 {
255         struct lowcore *lc = pcpu->lowcore;
256
257         cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
258         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
259         lc->cpu_nr = cpu;
260         lc->spinlock_lockval = arch_spin_lockval(cpu);
261         lc->spinlock_index = 0;
262         lc->percpu_offset = __per_cpu_offset[cpu];
263         lc->kernel_asce = S390_lowcore.kernel_asce;
264         lc->user_asce = S390_lowcore.kernel_asce;
265         lc->machine_flags = S390_lowcore.machine_flags;
266         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
267         __ctl_store(lc->cregs_save_area, 0, 15);
268         lc->cregs_save_area[1] = lc->kernel_asce;
269         lc->cregs_save_area[7] = lc->vdso_asce;
270         save_access_regs((unsigned int *) lc->access_regs_save_area);
271         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
272                sizeof(lc->stfle_fac_list));
273         memcpy(lc->alt_stfle_fac_list, S390_lowcore.alt_stfle_fac_list,
274                sizeof(lc->alt_stfle_fac_list));
275         arch_spin_lock_setup(cpu);
276 }
277
278 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
279 {
280         struct lowcore *lc = pcpu->lowcore;
281
282         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
283                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
284         lc->current_task = (unsigned long) tsk;
285         lc->lpp = LPP_MAGIC;
286         lc->current_pid = tsk->pid;
287         lc->user_timer = tsk->thread.user_timer;
288         lc->guest_timer = tsk->thread.guest_timer;
289         lc->system_timer = tsk->thread.system_timer;
290         lc->hardirq_timer = tsk->thread.hardirq_timer;
291         lc->softirq_timer = tsk->thread.softirq_timer;
292         lc->steal_timer = 0;
293 }
294
295 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
296 {
297         struct lowcore *lc = pcpu->lowcore;
298
299         lc->restart_stack = lc->kernel_stack;
300         lc->restart_fn = (unsigned long) func;
301         lc->restart_data = (unsigned long) data;
302         lc->restart_source = -1UL;
303         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
304 }
305
306 /*
307  * Call function via PSW restart on pcpu and stop the current cpu.
308  */
309 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
310                           void *data, unsigned long stack)
311 {
312         struct lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
313         unsigned long source_cpu = stap();
314
315         __load_psw_mask(PSW_KERNEL_BITS);
316         if (pcpu->address == source_cpu)
317                 func(data);     /* should not return */
318         /* Stop target cpu (if func returns this stops the current cpu). */
319         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
320         /* Restart func on the target cpu and stop the current cpu. */
321         mem_assign_absolute(lc->restart_stack, stack);
322         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
323         mem_assign_absolute(lc->restart_data, (unsigned long) data);
324         mem_assign_absolute(lc->restart_source, source_cpu);
325         __bpon();
326         asm volatile(
327                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
328                 "       brc     2,0b    # busy, try again\n"
329                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
330                 "       brc     2,1b    # busy, try again\n"
331                 : : "d" (pcpu->address), "d" (source_cpu),
332                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
333                 : "0", "1", "cc");
334         for (;;) ;
335 }
336
337 /*
338  * Enable additional logical cpus for multi-threading.
339  */
340 static int pcpu_set_smt(unsigned int mtid)
341 {
342         int cc;
343
344         if (smp_cpu_mtid == mtid)
345                 return 0;
346         cc = __pcpu_sigp(0, SIGP_SET_MULTI_THREADING, mtid, NULL);
347         if (cc == 0) {
348                 smp_cpu_mtid = mtid;
349                 smp_cpu_mt_shift = 0;
350                 while (smp_cpu_mtid >= (1U << smp_cpu_mt_shift))
351                         smp_cpu_mt_shift++;
352                 pcpu_devices[0].address = stap();
353         }
354         return cc;
355 }
356
357 /*
358  * Call function on an online CPU.
359  */
360 void smp_call_online_cpu(void (*func)(void *), void *data)
361 {
362         struct pcpu *pcpu;
363
364         /* Use the current cpu if it is online. */
365         pcpu = pcpu_find_address(cpu_online_mask, stap());
366         if (!pcpu)
367                 /* Use the first online cpu. */
368                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
369         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
370 }
371
372 /*
373  * Call function on the ipl CPU.
374  */
375 void smp_call_ipl_cpu(void (*func)(void *), void *data)
376 {
377         struct lowcore *lc = pcpu_devices->lowcore;
378
379         if (pcpu_devices[0].address == stap())
380                 lc = &S390_lowcore;
381
382         pcpu_delegate(&pcpu_devices[0], func, data,
383                       lc->panic_stack - PANIC_FRAME_OFFSET + PAGE_SIZE);
384 }
385
386 int smp_find_processor_id(u16 address)
387 {
388         int cpu;
389
390         for_each_present_cpu(cpu)
391                 if (pcpu_devices[cpu].address == address)
392                         return cpu;
393         return -1;
394 }
395
396 bool notrace arch_vcpu_is_preempted(int cpu)
397 {
398         if (test_cpu_flag_of(CIF_ENABLED_WAIT, cpu))
399                 return false;
400         if (pcpu_running(pcpu_devices + cpu))
401                 return false;
402         return true;
403 }
404 EXPORT_SYMBOL(arch_vcpu_is_preempted);
405
406 void notrace smp_yield_cpu(int cpu)
407 {
408         if (MACHINE_HAS_DIAG9C) {
409                 diag_stat_inc_norecursion(DIAG_STAT_X09C);
410                 asm volatile("diag %0,0,0x9c"
411                              : : "d" (pcpu_devices[cpu].address));
412         } else if (MACHINE_HAS_DIAG44) {
413                 diag_stat_inc_norecursion(DIAG_STAT_X044);
414                 asm volatile("diag 0,0,0x44");
415         }
416 }
417
418 /*
419  * Send cpus emergency shutdown signal. This gives the cpus the
420  * opportunity to complete outstanding interrupts.
421  */
422 void notrace smp_emergency_stop(void)
423 {
424         cpumask_t cpumask;
425         u64 end;
426         int cpu;
427
428         cpumask_copy(&cpumask, cpu_online_mask);
429         cpumask_clear_cpu(smp_processor_id(), &cpumask);
430
431         end = get_tod_clock() + (1000000UL << 12);
432         for_each_cpu(cpu, &cpumask) {
433                 struct pcpu *pcpu = pcpu_devices + cpu;
434                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
435                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
436                                    0, NULL) == SIGP_CC_BUSY &&
437                        get_tod_clock() < end)
438                         cpu_relax();
439         }
440         while (get_tod_clock() < end) {
441                 for_each_cpu(cpu, &cpumask)
442                         if (pcpu_stopped(pcpu_devices + cpu))
443                                 cpumask_clear_cpu(cpu, &cpumask);
444                 if (cpumask_empty(&cpumask))
445                         break;
446                 cpu_relax();
447         }
448 }
449 NOKPROBE_SYMBOL(smp_emergency_stop);
450
451 /*
452  * Stop all cpus but the current one.
453  */
454 void smp_send_stop(void)
455 {
456         int cpu;
457
458         /* Disable all interrupts/machine checks */
459         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
460         trace_hardirqs_off();
461
462         debug_set_critical();
463
464         if (oops_in_progress)
465                 smp_emergency_stop();
466
467         /* stop all processors */
468         for_each_online_cpu(cpu) {
469                 if (cpu == smp_processor_id())
470                         continue;
471                 pcpu_sigp_retry(pcpu_devices + cpu, SIGP_STOP, 0);
472                 while (!pcpu_stopped(pcpu_devices + cpu))
473                         cpu_relax();
474         }
475 }
476
477 /*
478  * This is the main routine where commands issued by other
479  * cpus are handled.
480  */
481 static void smp_handle_ext_call(void)
482 {
483         unsigned long bits;
484
485         /* handle bit signal external calls */
486         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
487         if (test_bit(ec_stop_cpu, &bits))
488                 smp_stop_cpu();
489         if (test_bit(ec_schedule, &bits))
490                 scheduler_ipi();
491         if (test_bit(ec_call_function_single, &bits))
492                 generic_smp_call_function_single_interrupt();
493 }
494
495 static void do_ext_call_interrupt(struct ext_code ext_code,
496                                   unsigned int param32, unsigned long param64)
497 {
498         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
499         smp_handle_ext_call();
500 }
501
502 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
503 {
504         int cpu;
505
506         for_each_cpu(cpu, mask)
507                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
508 }
509
510 void arch_send_call_function_single_ipi(int cpu)
511 {
512         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
513 }
514
515 /*
516  * this function sends a 'reschedule' IPI to another CPU.
517  * it goes straight through and wastes no time serializing
518  * anything. Worst case is that we lose a reschedule ...
519  */
520 void smp_send_reschedule(int cpu)
521 {
522         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
523 }
524
525 /*
526  * parameter area for the set/clear control bit callbacks
527  */
528 struct ec_creg_mask_parms {
529         unsigned long orval;
530         unsigned long andval;
531         int cr;
532 };
533
534 /*
535  * callback for setting/clearing control bits
536  */
537 static void smp_ctl_bit_callback(void *info)
538 {
539         struct ec_creg_mask_parms *pp = info;
540         unsigned long cregs[16];
541
542         __ctl_store(cregs, 0, 15);
543         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
544         __ctl_load(cregs, 0, 15);
545 }
546
547 /*
548  * Set a bit in a control register of all cpus
549  */
550 void smp_ctl_set_bit(int cr, int bit)
551 {
552         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
553
554         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
555 }
556 EXPORT_SYMBOL(smp_ctl_set_bit);
557
558 /*
559  * Clear a bit in a control register of all cpus
560  */
561 void smp_ctl_clear_bit(int cr, int bit)
562 {
563         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
564
565         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
566 }
567 EXPORT_SYMBOL(smp_ctl_clear_bit);
568
569 #ifdef CONFIG_CRASH_DUMP
570
571 int smp_store_status(int cpu)
572 {
573         struct pcpu *pcpu = pcpu_devices + cpu;
574         unsigned long pa;
575
576         pa = __pa(&pcpu->lowcore->floating_pt_save_area);
577         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_STATUS_AT_ADDRESS,
578                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
579                 return -EIO;
580         if (!MACHINE_HAS_VX && !MACHINE_HAS_GS)
581                 return 0;
582         pa = __pa(pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK);
583         if (MACHINE_HAS_GS)
584                 pa |= pcpu->lowcore->mcesad & MCESA_LC_MASK;
585         if (__pcpu_sigp_relax(pcpu->address, SIGP_STORE_ADDITIONAL_STATUS,
586                               pa) != SIGP_CC_ORDER_CODE_ACCEPTED)
587                 return -EIO;
588         return 0;
589 }
590
591 /*
592  * Collect CPU state of the previous, crashed system.
593  * There are four cases:
594  * 1) standard zfcp dump
595  *    condition: OLDMEM_BASE == NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
596  *    The state for all CPUs except the boot CPU needs to be collected
597  *    with sigp stop-and-store-status. The boot CPU state is located in
598  *    the absolute lowcore of the memory stored in the HSA. The zcore code
599  *    will copy the boot CPU state from the HSA.
600  * 2) stand-alone kdump for SCSI (zfcp dump with swapped memory)
601  *    condition: OLDMEM_BASE != NULL && ipl_info.type == IPL_TYPE_FCP_DUMP
602  *    The state for all CPUs except the boot CPU needs to be collected
603  *    with sigp stop-and-store-status. The firmware or the boot-loader
604  *    stored the registers of the boot CPU in the absolute lowcore in the
605  *    memory of the old system.
606  * 3) kdump and the old kernel did not store the CPU state,
607  *    or stand-alone kdump for DASD
608  *    condition: OLDMEM_BASE != NULL && !is_kdump_kernel()
609  *    The state for all CPUs except the boot CPU needs to be collected
610  *    with sigp stop-and-store-status. The kexec code or the boot-loader
611  *    stored the registers of the boot CPU in the memory of the old system.
612  * 4) kdump and the old kernel stored the CPU state
613  *    condition: OLDMEM_BASE != NULL && is_kdump_kernel()
614  *    This case does not exist for s390 anymore, setup_arch explicitly
615  *    deactivates the elfcorehdr= kernel parameter
616  */
617 static __init void smp_save_cpu_vxrs(struct save_area *sa, u16 addr,
618                                      bool is_boot_cpu, unsigned long page)
619 {
620         __vector128 *vxrs = (__vector128 *) page;
621
622         if (is_boot_cpu)
623                 vxrs = boot_cpu_vector_save_area;
624         else
625                 __pcpu_sigp_relax(addr, SIGP_STORE_ADDITIONAL_STATUS, page);
626         save_area_add_vxrs(sa, vxrs);
627 }
628
629 static __init void smp_save_cpu_regs(struct save_area *sa, u16 addr,
630                                      bool is_boot_cpu, unsigned long page)
631 {
632         void *regs = (void *) page;
633
634         if (is_boot_cpu)
635                 copy_oldmem_kernel(regs, (void *) __LC_FPREGS_SAVE_AREA, 512);
636         else
637                 __pcpu_sigp_relax(addr, SIGP_STORE_STATUS_AT_ADDRESS, page);
638         save_area_add_regs(sa, regs);
639 }
640
641 void __init smp_save_dump_cpus(void)
642 {
643         int addr, boot_cpu_addr, max_cpu_addr;
644         struct save_area *sa;
645         unsigned long page;
646         bool is_boot_cpu;
647
648         if (!(OLDMEM_BASE || ipl_info.type == IPL_TYPE_FCP_DUMP))
649                 /* No previous system present, normal boot. */
650                 return;
651         /* Allocate a page as dumping area for the store status sigps */
652         page = memblock_alloc_base(PAGE_SIZE, PAGE_SIZE, 1UL << 31);
653         /* Set multi-threading state to the previous system. */
654         pcpu_set_smt(sclp.mtid_prev);
655         boot_cpu_addr = stap();
656         max_cpu_addr = SCLP_MAX_CORES << sclp.mtid_prev;
657         for (addr = 0; addr <= max_cpu_addr; addr++) {
658                 if (__pcpu_sigp_relax(addr, SIGP_SENSE, 0) ==
659                     SIGP_CC_NOT_OPERATIONAL)
660                         continue;
661                 is_boot_cpu = (addr == boot_cpu_addr);
662                 /* Allocate save area */
663                 sa = save_area_alloc(is_boot_cpu);
664                 if (!sa)
665                         panic("could not allocate memory for save area\n");
666                 if (MACHINE_HAS_VX)
667                         /* Get the vector registers */
668                         smp_save_cpu_vxrs(sa, addr, is_boot_cpu, page);
669                 /*
670                  * For a zfcp dump OLDMEM_BASE == NULL and the registers
671                  * of the boot CPU are stored in the HSA. To retrieve
672                  * these registers an SCLP request is required which is
673                  * done by drivers/s390/char/zcore.c:init_cpu_info()
674                  */
675                 if (!is_boot_cpu || OLDMEM_BASE)
676                         /* Get the CPU registers */
677                         smp_save_cpu_regs(sa, addr, is_boot_cpu, page);
678         }
679         memblock_free(page, PAGE_SIZE);
680         diag308_reset();
681         pcpu_set_smt(0);
682 }
683 #endif /* CONFIG_CRASH_DUMP */
684
685 void smp_cpu_set_polarization(int cpu, int val)
686 {
687         pcpu_devices[cpu].polarization = val;
688 }
689
690 int smp_cpu_get_polarization(int cpu)
691 {
692         return pcpu_devices[cpu].polarization;
693 }
694
695 static void __ref smp_get_core_info(struct sclp_core_info *info, int early)
696 {
697         static int use_sigp_detection;
698         int address;
699
700         if (use_sigp_detection || sclp_get_core_info(info, early)) {
701                 use_sigp_detection = 1;
702                 for (address = 0;
703                      address < (SCLP_MAX_CORES << smp_cpu_mt_shift);
704                      address += (1U << smp_cpu_mt_shift)) {
705                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0) ==
706                             SIGP_CC_NOT_OPERATIONAL)
707                                 continue;
708                         info->core[info->configured].core_id =
709                                 address >> smp_cpu_mt_shift;
710                         info->configured++;
711                 }
712                 info->combined = info->configured;
713         }
714 }
715
716 static int smp_add_present_cpu(int cpu);
717
718 static int smp_add_core(struct sclp_core_entry *core, cpumask_t *avail,
719                         bool configured, bool early)
720 {
721         struct pcpu *pcpu;
722         int cpu, nr, i;
723         u16 address;
724
725         nr = 0;
726         if (sclp.has_core_type && core->type != boot_core_type)
727                 return nr;
728         cpu = cpumask_first(avail);
729         address = core->core_id << smp_cpu_mt_shift;
730         for (i = 0; (i <= smp_cpu_mtid) && (cpu < nr_cpu_ids); i++) {
731                 if (pcpu_find_address(cpu_present_mask, address + i))
732                         continue;
733                 pcpu = pcpu_devices + cpu;
734                 pcpu->address = address + i;
735                 if (configured)
736                         pcpu->state = CPU_STATE_CONFIGURED;
737                 else
738                         pcpu->state = CPU_STATE_STANDBY;
739                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
740                 set_cpu_present(cpu, true);
741                 if (!early && smp_add_present_cpu(cpu) != 0)
742                         set_cpu_present(cpu, false);
743                 else
744                         nr++;
745                 cpumask_clear_cpu(cpu, avail);
746                 cpu = cpumask_next(cpu, avail);
747         }
748         return nr;
749 }
750
751 static int __smp_rescan_cpus(struct sclp_core_info *info, bool early)
752 {
753         struct sclp_core_entry *core;
754         static cpumask_t avail;
755         bool configured;
756         u16 core_id;
757         int nr, i;
758
759         nr = 0;
760         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
761         /*
762          * Add IPL core first (which got logical CPU number 0) to make sure
763          * that all SMT threads get subsequent logical CPU numbers.
764          */
765         if (early) {
766                 core_id = pcpu_devices[0].address >> smp_cpu_mt_shift;
767                 for (i = 0; i < info->configured; i++) {
768                         core = &info->core[i];
769                         if (core->core_id == core_id) {
770                                 nr += smp_add_core(core, &avail, true, early);
771                                 break;
772                         }
773                 }
774         }
775         for (i = 0; i < info->combined; i++) {
776                 configured = i < info->configured;
777                 nr += smp_add_core(&info->core[i], &avail, configured, early);
778         }
779         return nr;
780 }
781
782 void __init smp_detect_cpus(void)
783 {
784         unsigned int cpu, mtid, c_cpus, s_cpus;
785         struct sclp_core_info *info;
786         u16 address;
787
788         /* Get CPU information */
789         info = memblock_virt_alloc(sizeof(*info), 8);
790         smp_get_core_info(info, 1);
791         /* Find boot CPU type */
792         if (sclp.has_core_type) {
793                 address = stap();
794                 for (cpu = 0; cpu < info->combined; cpu++)
795                         if (info->core[cpu].core_id == address) {
796                                 /* The boot cpu dictates the cpu type. */
797                                 boot_core_type = info->core[cpu].type;
798                                 break;
799                         }
800                 if (cpu >= info->combined)
801                         panic("Could not find boot CPU type");
802         }
803
804         /* Set multi-threading state for the current system */
805         mtid = boot_core_type ? sclp.mtid : sclp.mtid_cp;
806         mtid = (mtid < smp_max_threads) ? mtid : smp_max_threads - 1;
807         pcpu_set_smt(mtid);
808
809         /* Print number of CPUs */
810         c_cpus = s_cpus = 0;
811         for (cpu = 0; cpu < info->combined; cpu++) {
812                 if (sclp.has_core_type &&
813                     info->core[cpu].type != boot_core_type)
814                         continue;
815                 if (cpu < info->configured)
816                         c_cpus += smp_cpu_mtid + 1;
817                 else
818                         s_cpus += smp_cpu_mtid + 1;
819         }
820         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
821
822         /* Add CPUs present at boot */
823         get_online_cpus();
824         __smp_rescan_cpus(info, true);
825         put_online_cpus();
826         memblock_free_early((unsigned long)info, sizeof(*info));
827 }
828
829 /*
830  *      Activate a secondary processor.
831  */
832 static void smp_start_secondary(void *cpuvoid)
833 {
834         int cpu = raw_smp_processor_id();
835
836         S390_lowcore.last_update_clock = get_tod_clock();
837         S390_lowcore.restart_stack = (unsigned long) restart_stack;
838         S390_lowcore.restart_fn = (unsigned long) do_restart;
839         S390_lowcore.restart_data = 0;
840         S390_lowcore.restart_source = -1UL;
841         restore_access_regs(S390_lowcore.access_regs_save_area);
842         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
843         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
844         set_cpu_flag(CIF_ASCE_PRIMARY);
845         set_cpu_flag(CIF_ASCE_SECONDARY);
846         cpu_init();
847         rcu_cpu_starting(cpu);
848         preempt_disable();
849         init_cpu_timer();
850         vtime_init();
851         pfault_init();
852         notify_cpu_starting(cpu);
853         if (topology_cpu_dedicated(cpu))
854                 set_cpu_flag(CIF_DEDICATED_CPU);
855         else
856                 clear_cpu_flag(CIF_DEDICATED_CPU);
857         set_cpu_online(cpu, true);
858         inc_irq_stat(CPU_RST);
859         local_irq_enable();
860         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
861 }
862
863 /* Upping and downing of CPUs */
864 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
865 {
866         struct pcpu *pcpu = pcpu_devices + cpu;
867         int rc;
868
869         if (pcpu->state != CPU_STATE_CONFIGURED)
870                 return -EIO;
871         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
872             SIGP_CC_ORDER_CODE_ACCEPTED)
873                 return -EIO;
874
875         rc = pcpu_alloc_lowcore(pcpu, cpu);
876         if (rc)
877                 return rc;
878         pcpu_prepare_secondary(pcpu, cpu);
879         pcpu_attach_task(pcpu, tidle);
880         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
881         /* Wait until cpu puts itself in the online & active maps */
882         while (!cpu_online(cpu))
883                 cpu_relax();
884         return 0;
885 }
886
887 static unsigned int setup_possible_cpus __initdata;
888
889 static int __init _setup_possible_cpus(char *s)
890 {
891         get_option(&s, &setup_possible_cpus);
892         return 0;
893 }
894 early_param("possible_cpus", _setup_possible_cpus);
895
896 #ifdef CONFIG_HOTPLUG_CPU
897
898 int __cpu_disable(void)
899 {
900         unsigned long cregs[16];
901
902         /* Handle possible pending IPIs */
903         smp_handle_ext_call();
904         set_cpu_online(smp_processor_id(), false);
905         /* Disable pseudo page faults on this cpu. */
906         pfault_fini();
907         /* Disable interrupt sources via control register. */
908         __ctl_store(cregs, 0, 15);
909         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
910         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
911         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
912         __ctl_load(cregs, 0, 15);
913         clear_cpu_flag(CIF_NOHZ_DELAY);
914         return 0;
915 }
916
917 void __cpu_die(unsigned int cpu)
918 {
919         struct pcpu *pcpu;
920
921         /* Wait until target cpu is down */
922         pcpu = pcpu_devices + cpu;
923         while (!pcpu_stopped(pcpu))
924                 cpu_relax();
925         pcpu_free_lowcore(pcpu);
926         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
927         cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
928 }
929
930 void __noreturn cpu_die(void)
931 {
932         idle_task_exit();
933         __bpon();
934         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
935         for (;;) ;
936 }
937
938 #endif /* CONFIG_HOTPLUG_CPU */
939
940 void __init smp_fill_possible_mask(void)
941 {
942         unsigned int possible, sclp_max, cpu;
943
944         sclp_max = max(sclp.mtid, sclp.mtid_cp) + 1;
945         sclp_max = min(smp_max_threads, sclp_max);
946         sclp_max = (sclp.max_cores * sclp_max) ?: nr_cpu_ids;
947         possible = setup_possible_cpus ?: nr_cpu_ids;
948         possible = min(possible, sclp_max);
949         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
950                 set_cpu_possible(cpu, true);
951 }
952
953 void __init smp_prepare_cpus(unsigned int max_cpus)
954 {
955         /* request the 0x1201 emergency signal external interrupt */
956         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
957                 panic("Couldn't request external interrupt 0x1201");
958         /* request the 0x1202 external call external interrupt */
959         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
960                 panic("Couldn't request external interrupt 0x1202");
961 }
962
963 void __init smp_prepare_boot_cpu(void)
964 {
965         struct pcpu *pcpu = pcpu_devices;
966
967         WARN_ON(!cpu_present(0) || !cpu_online(0));
968         pcpu->state = CPU_STATE_CONFIGURED;
969         pcpu->lowcore = (struct lowcore *)(unsigned long) store_prefix();
970         S390_lowcore.percpu_offset = __per_cpu_offset[0];
971         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
972 }
973
974 void __init smp_cpus_done(unsigned int max_cpus)
975 {
976 }
977
978 void __init smp_setup_processor_id(void)
979 {
980         pcpu_devices[0].address = stap();
981         S390_lowcore.cpu_nr = 0;
982         S390_lowcore.spinlock_lockval = arch_spin_lockval(0);
983         S390_lowcore.spinlock_index = 0;
984 }
985
986 /*
987  * the frequency of the profiling timer can be changed
988  * by writing a multiplier value into /proc/profile.
989  *
990  * usually you want to run this on all CPUs ;)
991  */
992 int setup_profiling_timer(unsigned int multiplier)
993 {
994         return 0;
995 }
996
997 #ifdef CONFIG_HOTPLUG_CPU
998 static ssize_t cpu_configure_show(struct device *dev,
999                                   struct device_attribute *attr, char *buf)
1000 {
1001         ssize_t count;
1002
1003         mutex_lock(&smp_cpu_state_mutex);
1004         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
1005         mutex_unlock(&smp_cpu_state_mutex);
1006         return count;
1007 }
1008
1009 static ssize_t cpu_configure_store(struct device *dev,
1010                                    struct device_attribute *attr,
1011                                    const char *buf, size_t count)
1012 {
1013         struct pcpu *pcpu;
1014         int cpu, val, rc, i;
1015         char delim;
1016
1017         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1018                 return -EINVAL;
1019         if (val != 0 && val != 1)
1020                 return -EINVAL;
1021         get_online_cpus();
1022         mutex_lock(&smp_cpu_state_mutex);
1023         rc = -EBUSY;
1024         /* disallow configuration changes of online cpus and cpu 0 */
1025         cpu = dev->id;
1026         cpu = smp_get_base_cpu(cpu);
1027         if (cpu == 0)
1028                 goto out;
1029         for (i = 0; i <= smp_cpu_mtid; i++)
1030                 if (cpu_online(cpu + i))
1031                         goto out;
1032         pcpu = pcpu_devices + cpu;
1033         rc = 0;
1034         switch (val) {
1035         case 0:
1036                 if (pcpu->state != CPU_STATE_CONFIGURED)
1037                         break;
1038                 rc = sclp_core_deconfigure(pcpu->address >> smp_cpu_mt_shift);
1039                 if (rc)
1040                         break;
1041                 for (i = 0; i <= smp_cpu_mtid; i++) {
1042                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1043                                 continue;
1044                         pcpu[i].state = CPU_STATE_STANDBY;
1045                         smp_cpu_set_polarization(cpu + i,
1046                                                  POLARIZATION_UNKNOWN);
1047                 }
1048                 topology_expect_change();
1049                 break;
1050         case 1:
1051                 if (pcpu->state != CPU_STATE_STANDBY)
1052                         break;
1053                 rc = sclp_core_configure(pcpu->address >> smp_cpu_mt_shift);
1054                 if (rc)
1055                         break;
1056                 for (i = 0; i <= smp_cpu_mtid; i++) {
1057                         if (cpu + i >= nr_cpu_ids || !cpu_present(cpu + i))
1058                                 continue;
1059                         pcpu[i].state = CPU_STATE_CONFIGURED;
1060                         smp_cpu_set_polarization(cpu + i,
1061                                                  POLARIZATION_UNKNOWN);
1062                 }
1063                 topology_expect_change();
1064                 break;
1065         default:
1066                 break;
1067         }
1068 out:
1069         mutex_unlock(&smp_cpu_state_mutex);
1070         put_online_cpus();
1071         return rc ? rc : count;
1072 }
1073 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
1074 #endif /* CONFIG_HOTPLUG_CPU */
1075
1076 static ssize_t show_cpu_address(struct device *dev,
1077                                 struct device_attribute *attr, char *buf)
1078 {
1079         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
1080 }
1081 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
1082
1083 static struct attribute *cpu_common_attrs[] = {
1084 #ifdef CONFIG_HOTPLUG_CPU
1085         &dev_attr_configure.attr,
1086 #endif
1087         &dev_attr_address.attr,
1088         NULL,
1089 };
1090
1091 static struct attribute_group cpu_common_attr_group = {
1092         .attrs = cpu_common_attrs,
1093 };
1094
1095 static struct attribute *cpu_online_attrs[] = {
1096         &dev_attr_idle_count.attr,
1097         &dev_attr_idle_time_us.attr,
1098         NULL,
1099 };
1100
1101 static struct attribute_group cpu_online_attr_group = {
1102         .attrs = cpu_online_attrs,
1103 };
1104
1105 static int smp_cpu_online(unsigned int cpu)
1106 {
1107         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1108
1109         return sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1110 }
1111 static int smp_cpu_pre_down(unsigned int cpu)
1112 {
1113         struct device *s = &per_cpu(cpu_device, cpu)->dev;
1114
1115         sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1116         return 0;
1117 }
1118
1119 static int smp_add_present_cpu(int cpu)
1120 {
1121         struct device *s;
1122         struct cpu *c;
1123         int rc;
1124
1125         c = kzalloc(sizeof(*c), GFP_KERNEL);
1126         if (!c)
1127                 return -ENOMEM;
1128         per_cpu(cpu_device, cpu) = c;
1129         s = &c->dev;
1130         c->hotpluggable = 1;
1131         rc = register_cpu(c, cpu);
1132         if (rc)
1133                 goto out;
1134         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1135         if (rc)
1136                 goto out_cpu;
1137         rc = topology_cpu_init(c);
1138         if (rc)
1139                 goto out_topology;
1140         return 0;
1141
1142 out_topology:
1143         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1144 out_cpu:
1145 #ifdef CONFIG_HOTPLUG_CPU
1146         unregister_cpu(c);
1147 #endif
1148 out:
1149         return rc;
1150 }
1151
1152 #ifdef CONFIG_HOTPLUG_CPU
1153
1154 int __ref smp_rescan_cpus(void)
1155 {
1156         struct sclp_core_info *info;
1157         int nr;
1158
1159         info = kzalloc(sizeof(*info), GFP_KERNEL);
1160         if (!info)
1161                 return -ENOMEM;
1162         smp_get_core_info(info, 0);
1163         get_online_cpus();
1164         mutex_lock(&smp_cpu_state_mutex);
1165         nr = __smp_rescan_cpus(info, false);
1166         mutex_unlock(&smp_cpu_state_mutex);
1167         put_online_cpus();
1168         kfree(info);
1169         if (nr)
1170                 topology_schedule_update();
1171         return 0;
1172 }
1173
1174 static ssize_t __ref rescan_store(struct device *dev,
1175                                   struct device_attribute *attr,
1176                                   const char *buf,
1177                                   size_t count)
1178 {
1179         int rc;
1180
1181         rc = lock_device_hotplug_sysfs();
1182         if (rc)
1183                 return rc;
1184         rc = smp_rescan_cpus();
1185         unlock_device_hotplug();
1186         return rc ? rc : count;
1187 }
1188 static DEVICE_ATTR_WO(rescan);
1189 #endif /* CONFIG_HOTPLUG_CPU */
1190
1191 static int __init s390_smp_init(void)
1192 {
1193         int cpu, rc = 0;
1194
1195 #ifdef CONFIG_HOTPLUG_CPU
1196         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1197         if (rc)
1198                 return rc;
1199 #endif
1200         for_each_present_cpu(cpu) {
1201                 rc = smp_add_present_cpu(cpu);
1202                 if (rc)
1203                         goto out;
1204         }
1205
1206         rc = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "s390/smp:online",
1207                                smp_cpu_online, smp_cpu_pre_down);
1208         rc = rc <= 0 ? rc : 0;
1209 out:
1210         return rc;
1211 }
1212 subsys_initcall(s390_smp_init);